Extended Analysis of Binary Congestion Control

ABSTRACT Congestion ontrol in the Internet relies on binary adjustment algorithms. For example, Transmission Control Proto ol (TCP) in its ongestion avoidan e mode behaves similarly to Additive-In rease Multipli ative-De rease (AIMD) algorithm. The lassi al analysis by Chiu and Jain re ommends AIMD based on the assertion that among stable linear algorithms, AIMD ensures the qui kest onvergen e to fair states. We demonstrate in orre tness of this assertion. For an asyn hronous version of Chiu-Jain model, we show that AIMD is sensitive to initial onditions and has multiple unfair attra tors. Our ndings question the appropriateness of AIMD for binary ongestion ontrol. We attribute some of our observations to unrealisti features of Chiu-Jain model and argue that binary adjustment algorithms should be analyzed in a more realisti model.